WebMCP tool path
ForgeMCP includes runtime WebMCP detection, typed tool contracts and a real registration path using document.modelContext.registerTool(...) where supported.
CUBE CHESS 512 AI × FORGEMCP
WebMCP Challenge 2026
MULTI-AGENT RESEARCH & GAME STUDIO
Learn & Compete. Create. Verify.
ForgeMCP is the new open-source agent-native layer connecting humans and specialized AI agents to real application capabilities through WebMCP. Cube Chess 512 is one of its two existing laboratories. The deterministic 8×8×8 rules engine remains the source of truth for legal moves and game state.
HUMAN → GAME COORDINATOR → AI TRAINER / RULES ENGINEER / VISUAL AGENT / QA → WEBMCP TOOLS → CUBE RULE ENGINE → EXECUTED EXPERIMENT → VERIFICATION → HUMAN DECISION
First connected boundary
The current foundation includes deterministic Cube position inspection and integration health boundaries. Deeper rule-engine operations are the next integration stage.
Baseline vs candidate
Computer-vs-computer experiments must use executed legal games, measured results, regression checks and human approval before a candidate can be promoted.
Transparency: Cube Chess 512 existed before the WebMCP Challenge. ForgeMCP is new challenge work extending this existing game through controlled agent-facing WebMCP capabilities. No tournament, game count or AI improvement is treated as real unless it was actually executed and measured.
BASELINE AI vs CANDIDATE AI → legal games → stored moves/results → analyze errors → compare policies → retest → benchmark → legality/regression checks → human review → PROMOTE or REJECT
Self-play alone is not proof of improvement. Do not fabricate game counts, wins/losses, Elo, benchmarks or training results. The deterministic Cube rule engine remains the source of truth. The ForgeMCP WebMCP foundation and initial read-only Cube inspection boundary exist at framework level; the full self-play WebMCP tournament workflow is not yet fully connected. ForgeMCP does include its promotion gate and human approval model.